behavioral sciences Article Voluntary Language Switching in the Context of Bilingual Aphasia Nicholas Grunden 1 , Giorgio Piazza 2 , Carmen García-Sánchez 3 and Marco Calabria 4, * 1 Center for Brain and Cognition, Pompeu Fabra University, 08005 Barcelona, Spain; ngrunden3@gmail.com 2 Basque Center on Cognition, Brain and Language (BCBL), 20009 Donostia-San Sebastián, Spain; g.piazza@bcbl.eu 3 Hospital de la Santa Creu i Sant Pau, 08041 Barcelona, Spain; CGarciaS@santpau.cat 4 Faculty of Health Sciences, Universitat Oberta de Catalunya, 08018 Barcelona, Spain * Correspondence: mcalabria@uoc.edu; Tel.: +34-934-505-418  Received: 23 July 2020; Accepted: 14 September 2020; Published: 18 September 2020  Abstract: As studies of bilingual language control (BLC) seek to explore the underpinnings of bilinguals’ abilities to juggle two languages, different types of language switching tasks have been used to uncover switching and mixing effects and thereby reveal what proactive and reactive control mechanisms are involved in language switching. Voluntary language switching tasks, where a bilingual participant can switch freely between their languages while naming, are being utilized more often due to their greater ecological validity compared to cued switching paradigms. Because this type of task had not yet been applied to language switching in bilingual patients, our study sought to explore voluntary switching in bilinguals with aphasia (BWAs) as well as in healthy bilinguals. In Experiment 1, we replicated previously reported results of switch costs and mixing benefits within our own bilingual population of Catalan-Spanish bilinguals. With Experiment 2, we compared both the performances of BWAs as a group and as individuals against control group performance. Results illustrated a complex picture of language control abilities, indicating varying degrees of association and dissociation between factors of BLC. Given the diversity of impairments in BWAs’ language control mechanisms, we highlight the need to examine BLC at the individual level and through the lens of theoretical cognitive control frameworks in order to further parse out how bilinguals regulate their language switching. Keywords: bilingual aphasia; voluntary language switching; bilingual language control; proactive control; reactive control 1. Introduction Bilinguals have the uncanny ability to manage their languages. Once achieving moderate proficiency, a bilingual has the power to maintain a language throughout a conversation and avoid blurting out unwanted intrusions from their other languages. Then, given the need, many bilinguals can seamlessly “flip the script” and switch in and out of languages to communicate with the people around them. This set of abilities is usually termed as bilingual language control (BLC) and includes a number of cognitive processes [1–3]. Outlined in the Adaptive Control Hypothesis (ACH; [4,5]), a comprehensive description of BLC includes at least eight control processes: goal maintenance, conflict monitoring, interference suppression, salient cue detection, selective response inhibition, task engagement and disengagement, and opportunistic planning. In order to study how bilinguals effectively switch between languages, most studies thus far have employed experimental tasks with cued switching between languages (e.g., [6–13]; for recent reviews see [14,15]). In these tasks, subjects are explicitly shown what language they need to name a given Behav. Sci. 2020, 10, 141; doi:10.3390/bs10090141 www.mdpi.com/journal/behavsci Behav. Sci. 2020, 10, 141 2 of 23 stimulus in with a visual cue (e.g., a color, flag of a given language region, etc.) From the resulting data, it is possible to measure two types of cost, switch and mixing costs. Switch costs are calculated as the difference in naming latencies between “switch” trials, where the target language changed compared to the previous trial, and “repeat” or “stay” trials, where the naming language was the same as in the previous trial. Mixing costs are calculated as the difference between repeat and single trials, where single trials are those named in the non-mixed conditions, such that only one naming language is in use (e.g., [16,17]). However, in light of more recent findings of mixing benefits in some bilingual populations [18,19], it would be more accurate to refer to these general measures as switching effects and mixing effects. Conceptually, switching effects can be thought of as reflections of the ability to resolve cross-language interference, language engagement, and disengagement, while mixing effects are related to working memory mechanisms, such as the demand in maintaining task goals that are present in a dual-language situation [17,20,21]. Research with pathological populations performing these tasks has reported useful insights on how BLC works. For instance, Calabria et al. [22] administered a cued language switching task to a patient with pathological language switching, with results showing that the patient’s performance on the task was exactly the same as her performance in the more naturalistic connected speech condition. That is, the patient exhibited cross-language intrusions from their non-dominant into their dominant language in the switching task, just as they did when they were required to describe complex pictures or when they engaged in normal conversation. Furthermore, language switching tasks have been useful in assessing the integrity of BLC mechanisms. Calabria et al. [23] found that, in semantic dementia, switching abilities measured via language switching tasks may be spared despite a marked degradation of semantic memory and anomic deficits. Finally, a series of studies in patients with neurodegeneration in the basal ganglia have highlighted an increased impact on BLC deficits compared to other control mechanisms; results indicated that language switching abilities can be more affected than non-linguistic control abilities in bilingual patients with Parkinson’s disease (PD) [20] and that language switching is clearly dissociated from other language control abilities [21]. These studies thus reveal a crucial distinction between BLC deficits affecting control pathways in bilingual language production and generalized language deficits, as PD patients did not exhibit any type of language disorder. Given these findings, it is plausible then to hypothesize that patients with aphasia may also have some specific impairment in language switching abilities. There is evidence that, in some circumstances, bilingual patients with aphasia may show involuntary cross-language intrusions and/or language mixing by blending morphological features of the two languages within a word or words in a sentence [24–31]. More recently, we have also uncovered evidence that patients with aphasia without involuntary language mixing or switching have shown language control deficits at the lexical level that prevent them from performing effective word retrieval [32,33]. In bilinguals with aphasia (BWA) such as these who do not show pronounced pathological switching, language difficulties have been argued to stem from problems in controlling and managing the inhibition of their languages [34,35]. Therefore, it is reasonable to speculate that their control deficits may extend to difficulties in language disengagement and engagement, as well as language maintenance when they find themselves in contexts where switching back and forth between the two languages is required. However, little is currently known about these deficits in bilingual individuals with post-stroke aphasia that do not demonstrate overt involuntary language mixing or switching. Addressing this issue, we aimed to identify the BLC deficits that may prevent BWAs from efficiently engaging with their language switching abilities. To do so, we explored the performance of Catalan-Spanish BWA and healthy controls on a language switching task, specifically focusing on two key aspects. First, we used a voluntary language switching task as a relatively new experimental approach to this issue and a method of capturing a more natural switching behavior for our population of bilinguals. Voluntary switching is regarded as a more ecological measure of language switching [3], especially for bilinguals immersed in dual-language or dense code-switching contexts [4] such as is Behav. Sci. 2020, 10, 141 3 of 23 the case for most Catalan-Spanish bilinguals in the Barcelona metropolitan area. In past studies, this type of language switching task has been employed as an alternative to cued switching paradigms when studying the underlying control mechanisms of endogenous language switching within healthy individuals [18,19,36–38] but there is no study to date that incorporates patients with bilingual aphasia. Consequently, any findings in the voluntary switching task for BWA will hold clinical and ecological significance. Second, our study of language switching abilities in BWAs focused on two main control components measured via switching and mixing effects. In an ongoing debate surrounding the nature of BLC mechanisms, it has recently been proposed that these control mechanisms are differentially involved in single- vs. dual-language contexts of speech production (e.g., [16,17]). Furthermore, recent findings from cued language switching tasks in bilingual patients with PD have demonstrated that switch and mixing costs are possibly related to two qualitatively different mechanisms [20,21]. Within the context of the dual mechanisms of control (DMC) framework of non-linguistic executive control [39,40], these two dual-language effects have been associated with two different types of control, reactive control and proactive control. Reactive control, measured by switching effects, is defined as a bottom-up, transient, and stimulus-driven type of control whereas proactive control, measured by mixing effects, is top-down, more sustained, and goal-directed. This dual-mechanisms concept of control has subsequently been applied to explain the underlying mechanisms of BLC [17]: reactive control is engaged when bilinguals have to solve for cross-language interference in switch trials, while proactive control comes into play when they have to maintain their two languages active during a dual-language naming condition. Informing our current study, these two control mechanisms have been studied within the context of voluntary language switching with varying results. In some cases, it has been reported that participants did not show switch costs when utilizing certain voluntary naming strategies, suggesting an effective reliance upon reactive control (“bottom-up switching”; [41]). However, other experiments have shown the opposite, where participants still had switch costs in the voluntary switching condition [18,19]. In contrast to cued language switching, proactive control for bilinguals in voluntary language switching has largely revealed a mixing benefit, where bilinguals named pictures faster in mixed-language vs. single-language contexts [18,19,38]. Present Study Our study involved two experiments: Experiment 1, where we sought to replicate previous studies’ exploration of voluntary language switching [18] within our distinct bilingual population of Catalan-Spanish bilinguals; and Experiment 2, where we compared performance of BWAs and healthy bilingual controls on the voluntary language switching task. In both experiments, we tested participants with the following linguistic profile: early (learning both languages before 6 years old) and balanced (largely equivalent frequency of language usage) bilinguals with high levels of proficiency in both languages. Within Experiment 1, young Catalan-Spanish bilinguals participated in a picture naming task involving blocks with varying target languages: two single-language blocks (one in Spanish and one in Catalan) followed by a dual-language block where they were instructed to name the picture in either language. The design of the experiment is a replication of the de Bruin et al. [18] study, with the exception of cognate status in stimuli; in our study, half of the trials were cognate words (with phonological overlap between the Catalan and Spanish words for a given item) and half were non-cognate words (without phonological overlap). Because the modification is minor, we largely expected similar results to those obtained by de Bruin et al. [18]. First, we expected that the frequency of language switching would be around 50%, where about half of the trials in the dual-language conditions will be named in Catalan and about half in Spanish. Second, as was found by de Bruin et al. [18], we expected a mixing benefit for our bilinguals, measured by larger naming latencies for single trials (single-language condition) compared to repeat trials (dual-language condition). Third, despite this hypothesized benefit of mixing for bilinguals in the dual-language condition, we also expected to find Behav. Sci. 2020, 10, 141 4 of 23 a switch cost within the dual-language context. This would suggest that switching and the usage of language engagement and disengagement control processes would elicit a cost for the language system. The magnitude of the switch cost is expected to be the same for the two languages based on previous studies’ findings with high-proficient bilinguals in cued language switching [9,10,42,43] and by de Bruin et al. [18] in voluntary language switching. Finally, as an established body of literature maintains the presence of a cognate facilitation effect in healthy bilinguals [44,45], we likewise predict that our participants will benefit in cognate trials and demonstrate smaller naming latencies. Within Experiment 2, Catalan-Spanish bilinguals with aphasia and age-matched healthy controls completed a voluntary language switching task, as in Experiment 1. For the BWA group, we centered our predictions on the hypothesis that BWA may show deficits in monitoring their two languages during a dual-language naming condition. Evidence supporting this hypothesis comes from our previous study with bilingual aphasia, in which we looked at the relationship between semantic interference in picture naming and conflict monitoring [32]. Results showed that bilinguals with aphasia responded slower compared to healthy controls when performing an executive control task requiring them to monitor for both congruent and incongruent stimuli (a flanker task) but were as efficient as healthy controls when they had to solve for stimulus incongruence (conflict cost). Critically, BWA performance on the flanker task correlated with their delayed naming latencies on a picture naming task and was thus interpreted as evidence of a generalized conflict monitoring deficit. In the current experiment, deficits in monitoring could have repercussions when attempting to maintain two languages active in the dual-language condition and impact the proactive control system (for proactive non-linguistic deficits in bilingual patients with aphasia, see [46]). In light of these previous findings, our predictions for BWA compared to controls were the following: First, if BWAs have deficits that would greatly hinder their performance in the dual-language condition, we predicted that their switching frequency would be lower than 50%, as they presumably stick more to one language rather than alternating between the two. Second, if they do exhibit a lower rate of switching than controls, we predicted that the performance in the dual-language condition for BWAs would resemble the single-language condition and thus we would not expect to find a mixing benefit reported in previous studies (de Bruin et al., 2018). Third, we expected that patients with aphasia and healthy controls would show similar magnitudes of switch costs due to previous evidence that bilinguals with aphasia demonstrate preserved reactive control on a non-linguistic control task [46]. Finally, as has been established in healthy bilinguals as well as bilinguals with aphasia [47], we predicted that BWAs would benefit from a cognate facilitation effect. 2. Experiment 1: Young Adult Bilinguals 2.1. Methods 2.1.1. Participants A total of 20 university-age, healthy participants (12 women, 8 men) were recruited from a volunteer database at Pompeu Fabra University. Sociodemographic and language background information was collected via a pre-experimental questionnaire (see Table 1). All participants gave written informed consent in accordance with the Declaration of Helsinki. The experimental protocol was approved by the “Parc de Salut MAR” Research Ethics Committee (reference number: 2018/8029/I). Participants were compensated for their time during the experiment. Behav. Sci. 2020, 10, 141 5 of 23 Table 1. Sociodemographic and linguistic measures for young Catalan-Spanish bilinguals. M SD Age (years) 22.35 2.58 Education (years) 16.35 2.58 BSWQ Subscores (max. 15) DL Switch 8.65 1.84 NDL Switch 8.20 1.91 Contextual Switch 8.20 2.61 Unintended Switch 6.65 2.74 BSWQ Overall Switch (max. 60) 31.70 6.73 Dominant Language (DL) Age of Acquisition (years) 0.75 1.16 Proficiency (max. 7) Speaking 7.00 — Comprehension 6.75 0.64 Reading 6.85 0.49 Writing 7.00 — Language Usage (%) 54.97 14.48 Non-dominant Language (NDL) Age of acquisition (years) 1.75 1.74 Proficiency (max. 7) Speaking 7.00 — Comprehension 6.75 0.55 Reading 6.75 0.55 Writing 6.95 0.22 Language Usage (%) 34.19 14.10 In the questionnaire given to participants, age of acquisition and degrees of proficiency in reading, writing, speaking, and comprehension (rated on 7-point scales, with “1” indicating low proficiency and “7” indicating high proficiency) were recorded. Based on responses, subjects who took part in this study were all considered to be early, balanced Catalan-Spanish bilinguals, as indicated by their acquisition of both their languages before 5 years of age and having balanced proficiency in both languages. Additionally, because these measures did not indicate a clear L1 or L2, we labeled the participants’ languages as their dominant language (DL) and non-dominant language (NDL), according to self-reported dominance. In our sample, 10 participants were labeled as Catalan-dominant and the remaining 10 participants were Spanish-dominant. Frequency of usage of both Spanish and Catalan was reported as well. Finally, in order to obtain measures of switching behavior outside the experimental setting, participants’ switching was assessed using the Bilingual Switching Questionnaire (BSWQ; [48]). This questionnaire yielded 5 switching scores: L1-Switch, L2-Switch, Contextual Switching, Unintended Switching, and Overall Switching. Following our classification of our bilinguals’ two languages, L1-Switch and L2-Switch were termed DL-Switch and NDL-Switch, respectively. See Table 1 for sociodemographic and language information. 2.1.2. Procedure The visual stimuli for this task consisted of 60 different pictures taken from Snodgrass and Vanderwart [49]. Stimuli were balanced across Catalan and Spanish for measures of logarithmic frequency (Catalan: M = 1.19, SD = 0.52; Spanish: M = 1.11, SD = 0.46; t(59) = 0.90, p = 0.39), word length (Catalan: M = 5.86, SD = 1.44; Spanish: M = 5.70, SD = 1.25; t(59) = 0.66, p = 0.41), and number of phonemes (Catalan: M = 5.27, SD = 1.32; Spanish: M = 5.35, SD = 1.25; t(59) = 0.82, p = 0.91). The frequencies for the Spanish and Catalan names were obtained from the LEXESP [50] and the Catalan Dictionary of Frequencies [51] databases, respectively. In addition to these linguistic factors, half of the picture names were cognate words while the other half were non-cognate words between the two languages. Behav. Sci. 2020, 10, 141 6 of 23 At the onset of the experiment, subjects underwent a familiarization exercise where they were presented with all experimental pictures and asked to read aloud the correct responses for each picture printed below it in both Spanish and Catalan. While stimuli with high name agreement were selected for this experiment, this initial presentation of pictures served to strengthen this name agreement of stimuli across participants. Experimental tasks included three blocks: two single-language blocks and a dual-language block. Each single-language block consisted of 120 naming trials (60 pictures presented 2 times each) in either Spanish or Catalan. The order of these single-language blocks was counterbalanced across subjects and instructions for the task were always given in the target language of the block. These single-language blocks were then followed by a dual-language block (360 trials, 60 pictures repeated 6 times each), where participants were asked to name items with “whichever language comes most naturally” but also to “switch languages multiple times” throughout the task. The language of these instructions was counterbalanced across participants so as to balance any priming effects in subsequent naming. Dual-language blocks included two catch trials at the beginning of the picture naming which were then disregarded during analysis. Because of the longer duration of these blocks, subjects were given two opportunities to rest during the naming trials. Each of these breaks was also followed by another catch trial that was then taken out for analyses. For all trials regardless of block, subjects were presented with a fixation cross for 500 ms, followed by the image to be named by the participant. Said image remained on the screen for 2000 ms (irrespective of when a participant began/ended their response) and audio was recorded during this time to capture responses given. The picture naming task was administered to participants using DMDX software [52]. 2.1.3. Data Analysis Out of the 20 young bilingual participants recruited for this study, 1 participant was excluded due to an abnormally high number of errors (42 errors; group error average = 11.95). The remaining 19 participants were included in all subsequent statistical analyses. The dependent variables for this study, naming latencies and accuracy of participants’ responses, were analyzed off-line with CheckVocal [53]. For each audio file, the beginning of the participant’s response was marked and the response was coded as correct or incorrect. Naming latencies for incorrect responses and those exceeding 2 SDs above or below a given subject’s mean naming latency were excluded. Naming errors were classified under the following types [54]: (a) omissions: non-intelligible verbal response given during recording window or no response; (b) semantic errors: verbal response semantically related to target word; (c) formal errors: deletion, substitution, or addition of phonemes in target word; (d) unrelated errors: verbal response with no semantic or other relation to target word; (e) cross-language intrusions: correct naming of picture but in non-target language (only applicable for single-language blocks); and (f) auto-correction: incorrect verbal response followed by correction with target word. In order to examine effects of switching and mixing, trials in single-language blocks were classified as “single” trials whereas, in dual-language blocks, trials were labeled as “repeat” or “switch”, depending on whether the language used to name an item was the same (repeat) or different (switch) compared to the previous trial. For error trials that were able to be categorized as either Spanish or Catalan (semantic errors, non-ambiguous formal errors, unrelated errors and cross-language intrusions), the following trial was classified as repeat or switch trials in reference to the language of the error. However, for error trials that were not able to be categorized by language (i.e., no response and ambiguous formal errors), the trials directly after them were unclassifiable as repeat/switch and were excluded from analyses. Overall, an average of 4.58 trials (SD = 6.27), with a maximum of 25 trials, were unclassifiable per subject. With this data, we first analyzed the distribution of the switch trials per language and per cognate status in order to assess whether the switching frequency was around 50%, as done by de Bruin et al. [18]. Behav. Sci. 2020, 10, 141 7 of 23 Subsequently, we ran two separate repeated-measures ANOVAs for RTs and accuracy as dependent variables, including Trial Type (Single, Repeat, or Switch), Language (Dominant vs. Non-dominant) and Cognate Status (Cognate vs. Non-cognate) as within-subject factors. If the assumption of sphericity (Mauchly’s test) was violated, the Greenhouse–Geisser correction was applied by adjusting the degrees of freedom. 2.2. Results Switching frequencies. Participants on average switched on 43.99% (SD = 7.43) of the mixed language trials. Of those switch trials, there were similar percentages of switching for cognates Behav. Sci. 2020, 10, x 7 of 23 (M = 49.73, SD = 3.26) and non-cognates (M = 50.27, SD = 3.26; t(18) = −0.37, p = 0.718). Likewise, switch trials were equally shared between switches into dominant language (M = 50.31, SD = 0.97) and Naming latencies. Analyses showed that the main effect of Trial Type, F(1.17, 36) = 5.52, p = 0.02, switches into non-dominant language (M = 49.69, SD = 0.97; t(18) = 1.39, p = 0.182). ηp2 = 0.24, was significant. Post hoc analyses revealed a significant switch cost for participants, where Naming latencies. Analyses showed that the main effect of Trial Type, F(1.17, 36) = 5.52, p = 0.02, switch trials (M = 769 ms, SD = 96) were significantly slower than repeat trials (M = 753 ms, SD = 91; ηp 2 = 0.24, was significant. Post hoc analyses revealed a significant switch cost for participants, where p = 0.002). The difference in magnitude of switch cost was not significantly different across languages switch trials (M = 769 ms, SD = 96) were significantly slower than repeat trials (M = 753 ms, SD = 91; (DL switch cost: M = 2.36 ms, SD = 3.00; NDL switch cost: M = 2.07, SD = 3.79; t(18) = 0.245, p = 0.809). p = 0.002). The difference in magnitude of switch cost was not significantly different across languages Furthermore, naming for single trials (M = 788 ms, SD = 95) was significantly slower than for repeat (DL switch cost: M = 2.36 ms, SD = 3.00; NDL switch cost: M = 2.07, SD = 3.79; t(18) = 0.245, p = 0.809). trials (M = 753 ms, SD = 91; p = 0.04), suggesting a mixing benefit, where participants improved their Furthermore, naming for single trials (M = 788 ms, SD = 95) was significantly slower than for repeat naming latencies in dual-language conditions compared to single-language conditions (see Figure 1). trials (M = 753 ms, SD = 91; p = 0.04), suggesting a mixing benefit, where participants improved their naming latencies in dual-language conditions compared to single-language conditions (see Figure 1). Figure 1. Performance on voluntary switching task for young, healthy Catalan-Spanish bilinguals: (a) Figure Naming 1. latencies Performance showonsignificant voluntarydifferences switching task for young, between switchhealthy Catalan-Spanish and repeat trials (switchbilinguals: cost) and (a) Naming between latencies repeat show significant and single trials (mixingdifferences benefit) between with no switch effect ofand repeat trials language. (switch cost)ofand (b) Distribution between repeat proportional andand switching single trials mixing (mixing effects benefit) participants. in bilingual with no effect of language. (b) Distribution of proportional switching and mixing effects in bilingual participants. The main effect of Cognate Status was also significant, F(1,18) = 25.77, p < 0.001, ηp 2 = 0.59, where cognatesThe = 759effect (Mmain ms, SD = 88) were of Cognate named Status wasfaster than non-cognates also significant, F(1,18) =(M = 783p ms, 25.77, SD = < 0.001, ηp93). Neither 2 = 0.59, where thecognates main effect(M of = 759 ms, SD F(1,18) Language, = 0.18, = 88) were p = 0.68, named fasternor anynon-cognates than interactions between (M = 783 factors ms, SDwere= 93).found Neither to be thesignificant. main effect of Language, F(1,18) = 0.18, p = 0.68, nor any interactions between factors were found to Accuracy. Analyses revealed a significant main effect in Trial Type, F(1.51, 36) = 25.14, p < 0.001, be significant. ηp 2 = 0.58. Post hoc Accuracy. tests forrevealed Analyses Trial Type effects showed a significant maina effect significant decrease in Trial in accuracy Type, F(1.51, 36) =during 25.14, psingle < 0.001, trials (M = 97.24%, SD = 1.6) compared to both repeat (M = 99.12%, SD = 0.61; p < 0.001) ηp = 0.58. Post hoc tests for Trial Type effects showed a significant decrease in accuracy during single 2 and switch = 99.17%, (Mtrials SD = 0.78; (M = 97.24%, < 0.001) SD =p 1.6) trials.toThe compared bothmain effect repeat (M of CognateSD = 99.12%, Status = 0.61;wasp